AI in transition: From LLMs to agents

Prof. Dr Marko Robnik-Šikonja delivered a keynote speech at the MediaWiki Users and Developers Conference Fall 2026, exploring the key differences between large language models (LLMs) and agentic systems.

We are currently witnessing a shift in AI development from standalone large language models towards agentic systems. The keynote explained how both technologies work and what distinguishes them. Unlike traditional LLMs, agentic systems need to store and retrieve information, plan tasks and coordinate multiple actions. Wikis can serve as an important resource in supporting these processes.

The talk also introduced an agentic system designed to answer grammatical questions using empirically grounded evidence from the Universal Dependencies database, which covers more than 180 languages.

Watch the full keynote on YouTube. The talk starts at 57:00.